RESERVED GPU PRICING / B200 / B300 / GB300 NVL72
Reserved GPU pricing in 2026, by GPU and term
The lowest published 36-month rates are $3.79 per GPU-hour for a B200 and $4.70 for a B300. That is about 30% below the cheapest on-demand rate for each. Most providers quote reserved capacity privately. This page puts the published rates, our own rates and the monthly cost in one place.
Updated October 27, 2026. Prices checked October 17, 2026. SEC contracts read October 26, 2026. By GPU Rentals, powered by Bridgepointe Technologies. Every published figure comes from the provider's own pricing page. Sources are listed at the end.
Per GPU-hour, lowest published rate
Per GPU-hour, lowest published rate
Per GPU-hour, one provider's published discount
Our reserved rates by GPU and term
Budgetary rates based on real quotes we have received. Rates are per GPU-hour, billed monthly, with no upfront payment.
| Term | B200 | B300 | GB300 NVL72 |
|---|---|---|---|
| 12 months | $6.96 | $8.77 | Custom |
| 24 months | $5.41 | $6.79 | Custom |
| 36 months | $3.85 | $4.81 | Custom |
| 60 months | $3.47 | $4.33 | Custom |
Against AWS, Microsoft and Google
The three big clouds sell only 1-year and 3-year terms. Here is the lowest of their rates for each.
| GPU and term | Our rate | Lowest AWS, Microsoft or Google rate |
|---|---|---|
| B200, 12 months | $6.96 | $11.12, Google Cloud 1-year commitment |
| B200, 36 months | $3.85 | $6.15, AWS 3-year Savings Plan |
| B300, 12 months | $8.77 | $14.02, AWS 1-year Savings Plan |
| B300, 36 months | $4.81 | $7.69, AWS 3-year Savings Plan |
Every quote is custom. Your rate depends on GPU count, start date and region. GB300 NVL72 racks are priced deal by deal, so we quote them on request. Full detail by GPU: B200, B300, GB300 NVL72.
What a 64-GPU reserved cluster costs per month
Sixty-four GPUs is eight 8-GPU servers, a common size for a dedicated training cluster. These are our rates times 64 GPUs times 720 hours.
| Term | 64 B200s per month | 64 B300s per month |
|---|---|---|
| 12 months | $321,000 | $404,000 |
| 24 months | $249,000 | $313,000 |
| 36 months | $177,000 | $222,000 |
| 60 months | $160,000 | $200,000 |
Rounded to the nearest $1,000. Networking between servers, storage and data transfer are scoped to the workload and can add to these figures.
Published reserved GPU rates by provider
Only a few providers post a reserved rate. These are all the posted rates we found for each GPU, per GPU-hour.
| GPU | Provider | 12 months | 24 months | 36 months | Notes |
|---|---|---|---|---|---|
| B200 | Civo | $4.49 | $3.99 | $3.79 | 8-GPU instance. |
| B200 | Verda | $7.35 | $5.99 | Not published | Our math: published 8% and 25% term discounts applied to $7.99. |
| B200 | Google Cloud | $11.12 | Not offered | $7.09 | A4 VM, 8 GPUs, committed use discount. |
| B300 | Civo | $6.49 | $5.99 | $5.49 | 8-GPU node. |
| B300 | TheAI Cloud | Not offered | $5.00 | $4.70 | Whole 8-GPU nodes. 30% due at signing. |
| B300 | DigitalOcean | $7.94 | Not published | Not published | 12-month contract through sales. |
| B300 | Verda | $9.27 | $7.56 | Not published | Our math: published 8% and 25% term discounts applied to $10.08. |
| GB300 NVL72 | Verda | $10.66 | $8.69 | Not published | Our math: published 8% and 25% term discounts applied to $11.59. |
Nebius, CoreWeave, Runpod, Together AI, Hyperstack, Crusoe, Oracle, Microsoft Azure and Vultr rent these GPUs on reserved terms but quote the price privately. Nebius says reserved runs "up to 35% less" than on-demand. CoreWeave says committed use runs "up to 60%" less.
On-demand prices rose 33% to 56% at Lambda and Nebius in 2026. See the price trend charts on the B200 and B300 pages.
What large buyers sign: GPU contracts disclosed in SEC filings
Public GPU cloud companies have to report big contracts to the SEC. Six filings from 2026 state the GPU count, the term and the total contract value. Divide the value by GPUs and hours and you get the rate the buyer is paying. The companies did not publish these rates. We calculated them from their figures.
| Filing | Provider | Buyer | GPUs | Term | Contract value | Per GPU-hour, our math | Notes |
|---|---|---|---|---|---|---|---|
| 10-Q, Aug 2026 | WhiteFiber | Not named | 128 B300 | 36 months | $16.0 million | $4.76 | Closest in size to a typical cluster request. |
| 10-Q, Aug 2026 | WhiteFiber | BaseTen Labs | 1,392 B300 | 36 months | $165.2 million | $4.52 | |
| 8-K, May 2026 | Boost Run | Not named | 5,000 B300 | 36 months | $471.7 million | $3.59 | Value also covers storage and CPU servers. |
| 8-K, Apr 2026 | SharonAI | ESDS Software Solutions | 8,200 B300 | 60 months | $1.25 billion | $3.48 | In Australia. |
| 10-Q, Aug 2026 | WhiteFiber | Existing customer | 576 B300 | 60 months | $87.5 million | $3.47 | In Iceland. Revenue sharing on top. |
| 10-K, Mar 2026 | WhiteFiber | Existing customer | 384 B200 | 24 months | $18.1 million | $2.69 |
Our math: contract value divided by GPUs, months and 730 hours, the average month. With the 720-hour month used elsewhere on this page, each rate is about 1.4% higher. A contract value can include storage, networking and support, so the GPU-only rate may be lower. Filings report what the provider expects to earn, not a price list.
What this means for your quote
On 36-month terms, buyers of 128 to 5,000 B300s are paying $3.59 to $4.76 per GPU-hour. Our 36-month B300 rate is $4.81. The 128-GPU contract at $4.76 is the closest match in size, and it lands 5 cents below our rate. The lowest rates in these filings come with commitments most buyers do not make: thousands of GPUs, or five-year terms.
Who these providers are
- WhiteFiber (Nasdaq: WYFI). GPU cloud and data center company spun out of Bit Digital. Runs GPU clusters in Iceland and builds data centers in Canada and North Carolina.
- Boost Run (Nasdaq: BRUN). US GPU cloud, listed in May 2026. Runs its clusters in partner data centers at six US sites.
- SharonAI (Nasdaq: SHAZ). Australian GPU cloud, listed in February 2026. Runs its clusters in Equinix and NextDC data centers in Australia.
How much a reserved term saves
A reserved term commits you to a set number of GPUs for months or years. In return the hourly rate drops. Longer terms drop it further.
| GPU | Cheapest on-demand | Cheapest 12 months | Cheapest 36 months | 36 months vs on-demand |
|---|---|---|---|---|
| B200 | $5.43 | $4.49 | $3.79 | 30% lower |
| B300 | $6.60 | $6.49 | $4.70 | 29% lower |
| GB300 NVL72 | $11.59 | $10.66 | Not published | 24 months: 25% lower |
| GB200 NVL72, Microsoft Azure only | $27.04 | $17.31 | $11.90 | 56% lower |
For B200 and B300 the cheapest on-demand and the cheapest reserved rate come from different providers. The Azure row compares one provider's own rates for one VM, which is why its gap is larger.
Cluster size changes the price too
Lambda publishes B200 cluster rates by size, for terms of two weeks to one year: $9.86 per GPU-hour for 16 GPUs, $9.36 for 64 GPUs and $8.87 for 256 or more. B200 and B300 clusters grow in steps of 8, one server at a time. GB300 NVL72 clusters grow in steps of 72, one rack at a time.
What a reserved quote should spell out
A low hourly rate can hide costs elsewhere in the contract. Before you sign, check that the quote states each of these.
- The GPU count, term, start date and region.
- Prepayment. Some providers ask for part of the contract at signing.
- Cancellation and renewal terms.
- Networking between servers, storage and data transfer, which are often billed separately.
- Whether the provider has free capacity on your start date. A published price is not a promise of capacity.
That is the part we handle. Tell us the GPU, count, term and start date. We check which providers can actually deliver it and send back pricing.
Already in a reserved contract you no longer need? We can sublease or transfer the unused GPUs.
GET A RESERVED CLUSTER QUOTE
Reserved GPU clusters.
Checked for you.
Tell us which GPU, how many, for how long, and when. We check providers and reply by email with options and pricing. Nothing is booked or charged.
REQUEST SENT
Thanks. We have your request.
We will reply by email with provider options and pricing. Nothing has been booked or charged.
RESERVED GPU QUESTIONS
Before you reserve GPUs.
How much does a reserved GPU cost per hour?
Published 36-month rates start at $3.79 per GPU-hour for B200 and $4.70 for B300. Our own budgetary 36-month rates are $3.85 for B200 and $4.81 for B300, billed monthly.
How much cheaper is a reserved GPU than on-demand?
The cheapest published 36-month rates for B200 and B300 are about 30% below the cheapest on-demand rates. Within one provider the gap can be larger. Microsoft Azure's 3-year GB200 NVL72 rate is 56% below its pay-as-you-go rate.
What does a 64-GPU reserved cluster cost per month?
At our rates, about $321,000 a month for B200 and $404,000 for B300 on a 12-month term. On a 36-month term it is about $177,000 and $222,000. A month is 720 hours.
What reserved terms can I get?
Our rates cover 12, 24, 36 and 60 months. AWS, Microsoft and Google sell only 1-year and 3-year terms. Some providers also sell short terms. Together AI charges $6.79 per GPU-hour for B200 for 91 to 180 days.
Do I have to pay upfront for reserved GPUs?
Not with our rates, which are billed monthly with no upfront payment. Some providers ask for prepayment. TheAI Cloud asks for 30% at signing on its B300 nodes.
What do large companies pay for reserved GPUs?
SEC filings from 2026 show B300 contracts of 128 to 8,200 GPUs at $3.47 to $4.76 per GPU-hour on 3 to 5 year terms. We calculated those rates from each filing's GPU count, term and contract value. The providers did not publish them.
Why do most providers not publish reserved GPU prices?
Large commitments are negotiated one deal at a time. The price depends on GPU count, term, start date and region, and on whether the provider has free capacity on your start date.
Does a bigger cluster get a lower reserved rate?
Usually. Lambda publishes B200 cluster rates of $9.86 per GPU-hour for 16 GPUs, $9.36 for 64 GPUs and $8.87 for 256 or more.
Sources, checked October 17, 2026
Civo: civo.com/pricing; it lists three term prices per instance, read here as 12, 24 and 36 months in page order. Verda: verda.com/pricing. Google Cloud: cloud.google.com. TheAI Cloud: cloud.theai.com. DigitalOcean: digitalocean.com/pricing/gpu-droplets, effective August 1, 2026. Massed Compute: vm.massedcompute.com/pricing. Lambda: lambda.ai/pricing. Together AI: together.ai/pricing. Nebius: nebius.com/prices. CoreWeave: coreweave.com/pricing. Microsoft Azure retail prices API, ND GB200 v6, West US 3, effective April 1, 2025: prices.azure.com. AWS, Microsoft and Google comparison rates are detailed on our B200 and B300 pages.
SEC filings, read October 26, 2026. WhiteFiber 10-Q filed August 12, 2026: sec.gov. WhiteFiber 10-K filed March 26, 2026: sec.gov. Boost Run 8-K filed May 28, 2026: sec.gov. SharonAI 8-K filed April 1, 2026: sec.gov.
Prices change often and are shown as published. Monthly figures use 720 hours. A published price is not a promise that capacity is available.